\FloatBarrier
\begin{table*}[ht]
\centering
 \begin{tabular}{||c l l l ||}
 \hline
 Algorithms & Scenarios & Input & Output \\ [0.5ex]
\hline
 \hline
 Decision tree boosting \cite{decision tree} & Wireless networks & One-way delay, inter-packet times & Link loss or congestion loss\\[1ex]
 \hline
 Bayesian \cite{Bayesian} & Networks with reordered events & RTT of lost packets & Congestion or recording loss\\[1ex]
 \hline
 Hidden Markov model \cite{hidden-markov} & OBS & Number of bursts successfully received & Congestion or Contention loss\\
 & & at an egress between any two bursts &\\[1ex]
 \hline
 Multi-layer Perceptron \cite{Multi-layer Perceptron} & Wireless networks & Minimum RTT, current RTT & Congestion or wireless loss\\[1ex]
 \hline
 AdaBoost \cite{adaboost} & Satellite networks & Minimum RTT, the arrival time & Congestion or wireless loss\\
 & & interval of ACK &\\[1ex]
 \hline
 \hline
 \end{tabular}
 \caption{Existing Work in Supervised Learning}
\label{table:supervised}
\end{table*}
\FloatBarrier






\begin{table}[H]
\centering
 \begin{tabular}{||l||}
 \hline
 Objective\\ [0.5ex]
 \hline\hline
 Maximizing throughput \cite{Maximizing throughput}\\
 Minimizing RTT or flow completion time \cite{Minimizing RTT}\\
 Minimizing packet loss rate\cite{Minimizing packet}\\
 Fairness\cite{Fairness1}\cite{Fairness2}\\
 Responsiveness\cite{Responsiveness}\\ [1ex]
 \hline
 \end{tabular}
 \caption{Objectives of learning-based CC algorithms.}
\label{table:objectives}
\end{table}




\FloatBarrier
\begin{table*}[h]
\centering
 \begin{tabular}{||c c l l l||}
 \hline
 CC Algorithm Type & Algorithm & Details of the algorithm & Improved performance & Limitations \\ [0.5ex]
 \hline\hline
 Loss-based & Tahoe \cite{Tahoe} & Define the basic structure of CC & Fast network resource discovery  & Strains the network  \\
 & & algorithms including congestion, slow & along with high efficiency. & along with high-amplitude \\
 & & start avoidance, and fast retransmit. & & periodic phases.\\[1ex]
 \hline
 Loss-based & Reno \cite{Reno} & Add a fast recovery & Optimizes the slow start & Suffers performance degradation  \\
 & & & mechanism. & when the next packet losses \\
 & & & & and random losses exist.\\[1ex]
 \hline
 Loss-based & NewReno \cite{NewReno} & Modifies the fast recovery algorithm & Reduces packet losses and delay if & Limits the throughput due to the\\
 & & by introducing a response to partial & multiple packet losses occur & long-term recovery mechanism.\\
 & & acknowledgements. & together. & \\[1ex]
 \hline
 Delay-based & Vegas \cite{vegas} & Change the slow start mechanism to & Convergence speed and higher & Suffer low link utilization ratio\\
 & & control the number of extra buffers in & throughput. & compared with aggressive \\
 & & the network. & & algorithms such as Cubic \cite{cubic}.\\[1ex]
 \hline
 Delay-based & LoLa \cite{lola} & Show a fair flow balancing mechanism & Throughput, utilization and fairness. & Cannot coexist with loss-based\\
 & & for high-speed wide-area networks. & & CC algorithms fairly.\\[1ex]
 \hline
 Delay-based & FAST \cite{fast} & Show an equation-based algorithm and & Proportional fairness and not penalize & Available network resources\\
 & & use queuing delay as feedback to check & flows with large RTTs. & can't be allocated fairly.\\
 & & congestion. & &\\[1ex]
 \hline
 Hybrid-based & Veno \cite{veno} & Monitor congestion level based on RTT & Throughput. & Not address the issue of bursty\\
 & & packet losses and determine whether & & packet loss in wireless\\
 & & congestion occurs for the networks with & & networks.\\
 & & random packet loss. & &\\[1ex]
 \hline
 Hybrid-based & Africa \cite{africa} & Use delay information as an indicator & Throughput and fairness of high-BDP & Not been implemented in real\\
 & & towards an adaptive fair rapid increase & networks. & networks.\\
 & & for links with high BDP. & &\\[1ex]
 \hline
 Hybrid-based & Compound \cite{compound} & Propose a compound TCP, which adds a & Scalability and fairness in high-speed & Suffer fairness and latecomer\\
 & & delay-based component into Reno algorithm & and long-delay networks. & advantage issues when base\\
 & & for high-speed and long distant networks. & & RTT is wrongly estimated.\\[1ex]
 \hline
 \hline
 \end{tabular}
 \caption{End-to-End Congestion control algorithms}
\label{table:end-to-end}
\end{table*}
\FloatBarrier

\subsubsection{Network-assisted congestion control algorithms}
\begin{table*}[t]
\centering
 \begin{tabular}{||c l l l||}
 \hline
 Algorithm & Details of the algorithm & Improved performance & Limitations \\ [0.5ex]
 \hline\hline
 ECN \cite{ecn} & Allow end-to-end notification of CC to avoid & Avoid retransmission and reduce waiting & Performance depends on the accuracy of\\
 & packet loss, and require specific support from the & time, especially network jitter. & the AQM used.\\
 & Internet layer and the transport layer. & &\\[1ex]
 \hline
 QCN \cite{qcn} &
Include CP which is used to generate congestion & Ensure low queue buildup and low & Cannot be deployed and used in\\
& notification message and send to RP and RP is & queue oscillations. & large-scale IP networks since QCN\\
& responsible to deal with it and controls the & & cannot be used in IP routing networks.\\
& sending rate. & &\\[1ex]
 \hline
 \hline
 \end{tabular}
 \caption{Network Assisted Congestion control algorithms}
\label{table:network-assisted}
\end{table*}